🧪 AI Dual-Track Testing

One-command verification skill for Vibe Code projects.

Install

npx ai-dual-testing

That's it. Your AI tool now knows how to verify code.

Lock Requirements Upfront

To prevent scope drift, lock requirements with cryptographic SHA-256 integrity check before coding:

npx ai-dual-testing lock "1. User can register with email\n2. User can login"
# Or from a PRD file:
npx ai-dual-testing lock -f PRD.md

Usage

After install, tell your AI:

verify
verify feature login
kiểm tra
check coverage

The AI will:

• Lock / verify baseline requirements with SHA-256 checksum
• Safely generate & merge Playwright / Vitest test cases via test-writer.ts (anti-dummy test filter)
• Execute tests capturing real exit codes (no error swallowing)
• Generate timestamped RTM history and diff regressions
• Capture screenshots for UI evidence
• NOT auto-fix — you decide what to fix

Files created

.ai-testing/
├── scripts/
│   ├── verify.ts           ← Main verification runner (exit code & static audit enforced)
│   ├── test-writer.ts      ← Safe test creator & merger (anti-dummy test filter)
│   ├── test-auditor.ts     ← Static analysis engine (detects tautological/fake tests)
│   ├── master-rtm.ts       ← Timestamped RTM aggregator
│   ├── diff-rtm.ts         ← RTM historical regression & diff auditor
│   └── coverage-report.ts  ← Dual coverage check
├── configs/
│   ├── requirements.json   ← Baseline requirements (SHA-256 locked)
│   ├── thresholds.json     ← Coverage thresholds & static analysis rules
│   └── playwright.config.ts← Playwright E2E configuration
├── reports/
│   ├── screenshots/        ← Playwright screenshots
│   └── .gitkeep
└── e2e/                    ← Playwright test dir

Supported AI Tools

| Tool | Rule file | |------|-----------| | Cursor | .cursorrules (appended) | | Antigravity | .agents/skills/ai-testing/SKILL.md | | Claude Code | CLAUDE.md (appended) | | Windsurf | .windsurfrules (appended) | | None detected | AGENTS.md (created) |

Known Limitations

[!WARNING] Please be aware of the following structural limitations in the current version:

1. Filesystem Write Bypass: The AI Agent operates with your workspace permissions. If an agent deliberately bypasses the test-writer.ts script and directly calls filesystem tools on .spec.ts files, this cannot be blocked at the OS layer without a containerized sandbox.
2. Context Sharing (No Subagent Sandbox): Verification currently executes within the same chat session/context as the coding agent. Complete multi-agent role isolation (separating developer and QA into separate isolated processes) is planned for future versions.
3. Self-Referential / Circular Assertions: Static analysis detects literal tautologies (e.g. expect(true).toBe(true), expect(1).toBe(1), missing assertions, and hollow Playwright tests), but cannot statically detect circular/self-referential assertions (such as expect(fn()).toBe(fn()) calling the subject under test on both sides).

License

MIT